Segmentation fault when you release the encoder if your pip environment has scipy==1.15.2

[lantudou]

Version

0.4.1.21

Describe the bug.

My cuda version is 12.4. Here is a simple test example:

import cupy as cp
from nvidia import nvimgcodec


def test_encode():

    encoder = nvimgcodec.Encoder()

    
    test_image = cp.random.randint(0, 256, (1024, 1024, 1), dtype=cp.uint8)

    
    nv_image = nvimgcodec.as_image(test_image)
                
             
    encoder.write("test.jpg",
                    nv_image,
                    codec='jpeg')
                    
if __name__ == "__main__":
    test_encode()
    print("Process exiting...")  

My conda command:

conda create -n test3 python=3.12
pip install nvidia-nvimgcodec-cu12 cupy-cuda12x 

everything is ok, but when you install scipy==1.15.2, there is a Segmentation fault when the test program is exiting.

Minimum reproducible example

Environment details

Distributor ID: Ubuntu
Description:    Ubuntu 20.04.6 LTS
Release:        20.04
Codename:       focal

Relevant log output

python test.py
Process exiting...
Segmentation fault (core dumped)

Other/Misc.

No response

Check for duplicates

• I have searched the open bugs/issues and have found no duplicates for this bug report

[lantudou]

Downgrade scipy from 1.15.2 to 1.11.4, and make sure

import cupy as cp
from nvidia import nvimgcodec
import torch

rather than

import torch
import cupy as cp
from nvidia import nvimgcodec

could temporarily solve this problem, it is very strange thing.